Overclocking
I was rather eager to see just how far these modules would go given that they're rated at a rather strict 2-2-2-6-T1 DDR-400 SPD setting at default speeds; they seemed almost to be begging me to overclock them a little.
After ensuring that each module could operate at its stated speed by using the same basic hardware that's been present for all my memory-related reviews, namely a rather overclockable 2.26GHz Pentium 4 (3GHz+, on air, with a minimal voltage increase) and the tried and trusted ABIT IT7-MAX, I set both modules to my preferred running speed of 2-2-2-7-T1, as I've found the RAS-to-CAS setting to make minimal performance difference yet make a distinct difference in how far modules generally go. Upping the voltage to a reasonably safe 2.8v, this is how far the first module would go before bombing out on 3DMark 2001SE.
225MHz (DDR-450) @ 2-2-2-7-T1 is pretty impressive. I've always been an advocate of tight timings, and Corsair seem to have listened. The second module in this TWINX pack managed DDR-445 before submitting to exhaustion. Putting these two PC3200-rated modules in comparison with the others that I have tested previously, we see them in the following positions:
As you can see from the above graph, these latest low latency modules can be considered to be excellent performers at tight timings. That's hardly a surprise given their raison d'etre, but it's always nice to see product progression. The original XMS3200 C2 module, reviewed some months ago, now seems relatively pedestrian in comparison. It was anything but at the time of reviewing.
We've now ascertained that the modules are impressive performers when considered on a solitary basis. That answers one of our preset questions. How, now, do they perform as a pair, running alongside one another ?.
Using an XP2700 CPU coupled to a MSI nForce2-based motherboard gives us the chance to push the modules as far as we like. The various CPU-to-memory ratios allow us to keep the processor at its native 166FSB and simply up the memory speed.
The trouble with the nForce 2, though, and with the AMD scene in general is the lack of performance improvement as memory is run asynchronously. This is particularly evident with the dual-channel nForce 2, as it already produces far more bandwidth than the processor can utilise. Running memory asynchronously to the FSB in dual-channel mode is highlighted by a dip in performance when compared to running dual-channel memory in synchronous FSB mode.
Why is this ?. Simply because we already have 5.4GB/s of potential bandwidth at 166FSB; far more than the processor can use. Upping this amount to 6.4GB/s courtesy of running with dual-channel memory at DDR-400 speeds (PC3200) gives us more of what we can't use. Further, the buffering that's required to modulate the PC3200 memory speeds with the 166FSB CPU clock speed gives rise to unwanted latencies. In a nutshell, this PC3200LL matched memory will only hit specification if you run your CPU (unlocked, presumably) at 200FSB+ with synchronous memory speed.
Can it do PC3200 speeds, at 2-2-2-6-T1 timings in dual-channel mode ?. Sure it can. Running a 5:6 CPU:Memory ratio gave us the required memory speed although it cost us in the performance stakes. The twin TWINX modules never missed a beat at their rated timings. Indeed, the managed a reasonable 414MHz, in dual-channel mode, at 2-2-2-6-T1 before succumbing to failure. If your CPU is not unlocked and your nForce2 motherboard cannot hit 200FSB+ with ease, then you'll not be getting the best out of this matched pair. That same argument goes for the Intel Pentium 4 Granite Bay chipset, as it can only run memory synchronously to the CPU's FSB.